{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:3VDBKM6QQH7OPUXHLS57UEZZGT","short_pith_number":"pith:3VDBKM6Q","schema_version":"1.0","canonical_sha256":"dd461533d081fee7d2e75cbbfa133934fa3d7dc3e0bfd22394eead26e899e1b4","source":{"kind":"arxiv","id":"2412.04415","version":1},"attestation_state":"computed","paper":{"title":"Targeting the Core: A Simple and Effective Method to Attack RAG-based Agents via Direct LLM Manipulation","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Victor Bian, Xuying Li, Yasuhiro Yoshida, Yuji Kosuga, Zhuo Li","submitted_at":"2024-12-05T18:38:30Z","abstract_excerpt":"AI agents, powered by large language models (LLMs), have transformed human-computer interactions by enabling seamless, natural, and context-aware communication. While these advancements offer immense utility, they also inherit and amplify inherent safety risks such as bias, fairness, hallucinations, privacy breaches, and a lack of transparency. This paper investigates a critical vulnerability: adversarial attacks targeting the LLM core within AI agents. Specifically, we test the hypothesis that a deceptively simple adversarial prefix, such as \\textit{Ignore the document}, can compel LLMs to pr"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2412.04415","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/publicdomain/zero/1.0/","primary_cat":"cs.AI","submitted_at":"2024-12-05T18:38:30Z","cross_cats_sorted":[],"title_canon_sha256":"7c6f5f5412a132a920f77158a3d477abd68ee4342125c9c9aaf864e66aaa4dfe","abstract_canon_sha256":"0fd8a96ad697c1addc1d724f005ea2fcb86f21398f93d010233d04a2583f0829"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:45:03.646572Z","signature_b64":"bIPaGbcJU1Xv9Gdq1zv1tMXoI+Oo6V+rwNvKQrLZ7QHLES58cJa7xr3DTVqqBDm2StxPDmVFnNJgESQ1Gjp7Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dd461533d081fee7d2e75cbbfa133934fa3d7dc3e0bfd22394eead26e899e1b4","last_reissued_at":"2026-07-05T09:45:03.646116Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:45:03.646116Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Targeting the Core: A Simple and Effective Method to Attack RAG-based Agents via Direct LLM Manipulation","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Victor Bian, Xuying Li, Yasuhiro Yoshida, Yuji Kosuga, Zhuo Li","submitted_at":"2024-12-05T18:38:30Z","abstract_excerpt":"AI agents, powered by large language models (LLMs), have transformed human-computer interactions by enabling seamless, natural, and context-aware communication. While these advancements offer immense utility, they also inherit and amplify inherent safety risks such as bias, fairness, hallucinations, privacy breaches, and a lack of transparency. This paper investigates a critical vulnerability: adversarial attacks targeting the LLM core within AI agents. Specifically, we test the hypothesis that a deceptively simple adversarial prefix, such as \\textit{Ignore the document}, can compel LLMs to pr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.04415","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2412.04415/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2412.04415","created_at":"2026-07-05T09:45:03.646170+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.04415v1","created_at":"2026-07-05T09:45:03.646170+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.04415","created_at":"2026-07-05T09:45:03.646170+00:00"},{"alias_kind":"pith_short_12","alias_value":"3VDBKM6QQH7O","created_at":"2026-07-05T09:45:03.646170+00:00"},{"alias_kind":"pith_short_16","alias_value":"3VDBKM6QQH7OPUXH","created_at":"2026-07-05T09:45:03.646170+00:00"},{"alias_kind":"pith_short_8","alias_value":"3VDBKM6Q","created_at":"2026-07-05T09:45:03.646170+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12797","citing_title":"The Containment Gap: How Deployed Agentic AI Frameworks Fail Public-Facing Safety Requirements","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.10322","citing_title":"Game-Theoretic Multi-Agent Control for Robust Contextual Reasoning in LLMs","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2503.21460","citing_title":"Large Language Model Agent: A Survey on Methodology, Applications and Challenges","ref_index":204,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12813","citing_title":"REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations","ref_index":177,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3VDBKM6QQH7OPUXHLS57UEZZGT","json":"https://pith.science/pith/3VDBKM6QQH7OPUXHLS57UEZZGT.json","graph_json":"https://pith.science/api/pith-number/3VDBKM6QQH7OPUXHLS57UEZZGT/graph.json","events_json":"https://pith.science/api/pith-number/3VDBKM6QQH7OPUXHLS57UEZZGT/events.json","paper":"https://pith.science/paper/3VDBKM6Q"},"agent_actions":{"view_html":"https://pith.science/pith/3VDBKM6QQH7OPUXHLS57UEZZGT","download_json":"https://pith.science/pith/3VDBKM6QQH7OPUXHLS57UEZZGT.json","view_paper":"https://pith.science/paper/3VDBKM6Q","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.04415&json=true","fetch_graph":"https://pith.science/api/pith-number/3VDBKM6QQH7OPUXHLS57UEZZGT/graph.json","fetch_events":"https://pith.science/api/pith-number/3VDBKM6QQH7OPUXHLS57UEZZGT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3VDBKM6QQH7OPUXHLS57UEZZGT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3VDBKM6QQH7OPUXHLS57UEZZGT/action/storage_attestation","attest_author":"https://pith.science/pith/3VDBKM6QQH7OPUXHLS57UEZZGT/action/author_attestation","sign_citation":"https://pith.science/pith/3VDBKM6QQH7OPUXHLS57UEZZGT/action/citation_signature","submit_replication":"https://pith.science/pith/3VDBKM6QQH7OPUXHLS57UEZZGT/action/replication_record"}},"created_at":"2026-07-05T09:45:03.646170+00:00","updated_at":"2026-07-05T09:45:03.646170+00:00"}